PageRenderTime 79ms CodeModel.GetById 12ms app.highlight 59ms RepoModel.GetById 1ms app.codeStats 0ms

/src/FreeImage/Source/LibOpenJPEG/dwt.c

https://bitbucket.org/cabalistic/ogredeps/
C | 858 lines | 611 code | 98 blank | 149 comment | 117 complexity | 47d264d4de761375b5d99857967662e2 MD5 | raw file
  1/*
  2 * Copyright (c) 2002-2007, Communications and Remote Sensing Laboratory, Universite catholique de Louvain (UCL), Belgium
  3 * Copyright (c) 2002-2007, Professor Benoit Macq
  4 * Copyright (c) 2001-2003, David Janssens
  5 * Copyright (c) 2002-2003, Yannick Verschueren
  6 * Copyright (c) 2003-2007, Francois-Olivier Devaux and Antonin Descampe
  7 * Copyright (c) 2005, Herve Drolon, FreeImage Team
  8 * Copyright (c) 2007, Jonathan Ballard <dzonatas@dzonux.net>
  9 * Copyright (c) 2007, Callum Lerwick <seg@haxxed.com>
 10 * All rights reserved.
 11 *
 12 * Redistribution and use in source and binary forms, with or without
 13 * modification, are permitted provided that the following conditions
 14 * are met:
 15 * 1. Redistributions of source code must retain the above copyright
 16 *    notice, this list of conditions and the following disclaimer.
 17 * 2. Redistributions in binary form must reproduce the above copyright
 18 *    notice, this list of conditions and the following disclaimer in the
 19 *    documentation and/or other materials provided with the distribution.
 20 *
 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS `AS IS'
 22 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 24 * ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
 25 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 31 * POSSIBILITY OF SUCH DAMAGE.
 32 */
 33
 34#ifdef __SSE__
 35#include <xmmintrin.h>
 36#endif
 37
 38#include "opj_includes.h"
 39
 40/** @defgroup DWT DWT - Implementation of a discrete wavelet transform */
 41/*@{*/
 42
 43#define WS(i) v->mem[(i)*2]
 44#define WD(i) v->mem[(1+(i)*2)]
 45
 46/** @name Local data structures */
 47/*@{*/
 48
 49typedef struct dwt_local {
 50	int* mem;
 51	int dn;
 52	int sn;
 53	int cas;
 54} dwt_t;
 55
 56typedef union {
 57	float	f[4];
 58} v4;
 59
 60typedef struct v4dwt_local {
 61	v4*	wavelet ;
 62	int		dn ;
 63	int		sn ;
 64	int		cas ;
 65} v4dwt_t ;
 66
 67static const float dwt_alpha =  1.586134342f; /*  12994 */
 68static const float dwt_beta  =  0.052980118f; /*    434 */
 69static const float dwt_gamma = -0.882911075f; /*  -7233 */
 70static const float dwt_delta = -0.443506852f; /*  -3633 */
 71
 72static const float K      = 1.230174105f; /*  10078 */
 73/* FIXME: What is this constant? */
 74static const float c13318 = 1.625732422f;
 75
 76/*@}*/
 77
 78/**
 79Virtual function type for wavelet transform in 1-D 
 80*/
 81typedef void (*DWT1DFN)(dwt_t* v);
 82
 83/** @name Local static functions */
 84/*@{*/
 85
 86/**
 87Forward lazy transform (horizontal)
 88*/
 89static void dwt_deinterleave_h(int *a, int *b, int dn, int sn, int cas);
 90/**
 91Forward lazy transform (vertical)
 92*/
 93static void dwt_deinterleave_v(int *a, int *b, int dn, int sn, int x, int cas);
 94/**
 95Inverse lazy transform (horizontal)
 96*/
 97static void dwt_interleave_h(dwt_t* h, int *a);
 98/**
 99Inverse lazy transform (vertical)
100*/
101static void dwt_interleave_v(dwt_t* v, int *a, int x);
102/**
103Forward 5-3 wavelet transform in 1-D
104*/
105static void dwt_encode_1(int *a, int dn, int sn, int cas);
106/**
107Inverse 5-3 wavelet transform in 1-D
108*/
109static void dwt_decode_1(dwt_t *v);
110/**
111Forward 9-7 wavelet transform in 1-D
112*/
113static void dwt_encode_1_real(int *a, int dn, int sn, int cas);
114/**
115Explicit calculation of the Quantization Stepsizes 
116*/
117static void dwt_encode_stepsize(int stepsize, int numbps, opj_stepsize_t *bandno_stepsize);
118/**
119Inverse wavelet transform in 2-D.
120*/
121static void dwt_decode_tile(opj_tcd_tilecomp_t* tilec, int i, DWT1DFN fn);
122
123/*@}*/
124
125/*@}*/
126
127#define S(i) a[(i)*2]
128#define D(i) a[(1+(i)*2)]
129#define S_(i) ((i)<0?S(0):((i)>=sn?S(sn-1):S(i)))
130#define D_(i) ((i)<0?D(0):((i)>=dn?D(dn-1):D(i)))
131/* new */
132#define SS_(i) ((i)<0?S(0):((i)>=dn?S(dn-1):S(i)))
133#define DD_(i) ((i)<0?D(0):((i)>=sn?D(sn-1):D(i)))
134
135/* <summary>                                                              */
136/* This table contains the norms of the 5-3 wavelets for different bands. */
137/* </summary>                                                             */
138static const double dwt_norms[4][10] = {
139	{1.000, 1.500, 2.750, 5.375, 10.68, 21.34, 42.67, 85.33, 170.7, 341.3},
140	{1.038, 1.592, 2.919, 5.703, 11.33, 22.64, 45.25, 90.48, 180.9},
141	{1.038, 1.592, 2.919, 5.703, 11.33, 22.64, 45.25, 90.48, 180.9},
142	{.7186, .9218, 1.586, 3.043, 6.019, 12.01, 24.00, 47.97, 95.93}
143};
144
145/* <summary>                                                              */
146/* This table contains the norms of the 9-7 wavelets for different bands. */
147/* </summary>                                                             */
148static const double dwt_norms_real[4][10] = {
149	{1.000, 1.965, 4.177, 8.403, 16.90, 33.84, 67.69, 135.3, 270.6, 540.9},
150	{2.022, 3.989, 8.355, 17.04, 34.27, 68.63, 137.3, 274.6, 549.0},
151	{2.022, 3.989, 8.355, 17.04, 34.27, 68.63, 137.3, 274.6, 549.0},
152	{2.080, 3.865, 8.307, 17.18, 34.71, 69.59, 139.3, 278.6, 557.2}
153};
154
155/* 
156==========================================================
157   local functions
158==========================================================
159*/
160
161/* <summary>			                 */
162/* Forward lazy transform (horizontal).  */
163/* </summary>                            */ 
164static void dwt_deinterleave_h(int *a, int *b, int dn, int sn, int cas) {
165	int i;
166    for (i=0; i<sn; i++) b[i]=a[2*i+cas];
167    for (i=0; i<dn; i++) b[sn+i]=a[(2*i+1-cas)];
168}
169
170/* <summary>                             */  
171/* Forward lazy transform (vertical).    */
172/* </summary>                            */ 
173static void dwt_deinterleave_v(int *a, int *b, int dn, int sn, int x, int cas) {
174    int i;
175    for (i=0; i<sn; i++) b[i*x]=a[2*i+cas];
176    for (i=0; i<dn; i++) b[(sn+i)*x]=a[(2*i+1-cas)];
177}
178
179/* <summary>                             */
180/* Inverse lazy transform (horizontal).  */
181/* </summary>                            */
182static void dwt_interleave_h(dwt_t* h, int *a) {
183    int *ai = a;
184    int *bi = h->mem + h->cas;
185    int  i	= h->sn;
186    while( i-- ) {
187      *bi = *(ai++);
188	  bi += 2;
189    }
190    ai	= a + h->sn;
191    bi	= h->mem + 1 - h->cas;
192    i	= h->dn ;
193    while( i-- ) {
194      *bi = *(ai++);
195	  bi += 2;
196    }
197}
198
199/* <summary>                             */  
200/* Inverse lazy transform (vertical).    */
201/* </summary>                            */ 
202static void dwt_interleave_v(dwt_t* v, int *a, int x) {
203    int *ai = a;
204    int *bi = v->mem + v->cas;
205    int  i = v->sn;
206    while( i-- ) {
207      *bi = *ai;
208	  bi += 2;
209	  ai += x;
210    }
211    ai = a + (v->sn * x);
212    bi = v->mem + 1 - v->cas;
213    i = v->dn ;
214    while( i-- ) {
215      *bi = *ai;
216	  bi += 2;  
217	  ai += x;
218    }
219}
220
221
222/* <summary>                            */
223/* Forward 5-3 wavelet transform in 1-D. */
224/* </summary>                           */
225static void dwt_encode_1(int *a, int dn, int sn, int cas) {
226	int i;
227	
228	if (!cas) {
229		if ((dn > 0) || (sn > 1)) {	/* NEW :  CASE ONE ELEMENT */
230			for (i = 0; i < dn; i++) D(i) -= (S_(i) + S_(i + 1)) >> 1;
231			for (i = 0; i < sn; i++) S(i) += (D_(i - 1) + D_(i) + 2) >> 2;
232		}
233	} else {
234		if (!sn && dn == 1)		    /* NEW :  CASE ONE ELEMENT */
235			S(0) *= 2;
236		else {
237			for (i = 0; i < dn; i++) S(i) -= (DD_(i) + DD_(i - 1)) >> 1;
238			for (i = 0; i < sn; i++) D(i) += (SS_(i) + SS_(i + 1) + 2) >> 2;
239		}
240	}
241}
242
243/* <summary>                            */
244/* Inverse 5-3 wavelet transform in 1-D. */
245/* </summary>                           */ 
246static void dwt_decode_1_(int *a, int dn, int sn, int cas) {
247	int i;
248	
249	if (!cas) {
250		if ((dn > 0) || (sn > 1)) { /* NEW :  CASE ONE ELEMENT */
251			for (i = 0; i < sn; i++) S(i) -= (D_(i - 1) + D_(i) + 2) >> 2;
252			for (i = 0; i < dn; i++) D(i) += (S_(i) + S_(i + 1)) >> 1;
253		}
254	} else {
255		if (!sn  && dn == 1)          /* NEW :  CASE ONE ELEMENT */
256			S(0) /= 2;
257		else {
258			for (i = 0; i < sn; i++) D(i) -= (SS_(i) + SS_(i + 1) + 2) >> 2;
259			for (i = 0; i < dn; i++) S(i) += (DD_(i) + DD_(i - 1)) >> 1;
260		}
261	}
262}
263
264/* <summary>                            */
265/* Inverse 5-3 wavelet transform in 1-D. */
266/* </summary>                           */ 
267static void dwt_decode_1(dwt_t *v) {
268	dwt_decode_1_(v->mem, v->dn, v->sn, v->cas);
269}
270
271/* <summary>                             */
272/* Forward 9-7 wavelet transform in 1-D. */
273/* </summary>                            */
274static void dwt_encode_1_real(int *a, int dn, int sn, int cas) {
275	int i;
276	if (!cas) {
277		if ((dn > 0) || (sn > 1)) {	/* NEW :  CASE ONE ELEMENT */
278			for (i = 0; i < dn; i++)
279				D(i) -= fix_mul(S_(i) + S_(i + 1), 12993);
280			for (i = 0; i < sn; i++)
281				S(i) -= fix_mul(D_(i - 1) + D_(i), 434);
282			for (i = 0; i < dn; i++)
283				D(i) += fix_mul(S_(i) + S_(i + 1), 7233);
284			for (i = 0; i < sn; i++)
285				S(i) += fix_mul(D_(i - 1) + D_(i), 3633);
286			for (i = 0; i < dn; i++)
287				D(i) = fix_mul(D(i), 5038);	/*5038 */
288			for (i = 0; i < sn; i++)
289				S(i) = fix_mul(S(i), 6659);	/*6660 */
290		}
291	} else {
292		if ((sn > 0) || (dn > 1)) {	/* NEW :  CASE ONE ELEMENT */
293			for (i = 0; i < dn; i++)
294				S(i) -= fix_mul(DD_(i) + DD_(i - 1), 12993);
295			for (i = 0; i < sn; i++)
296				D(i) -= fix_mul(SS_(i) + SS_(i + 1), 434);
297			for (i = 0; i < dn; i++)
298				S(i) += fix_mul(DD_(i) + DD_(i - 1), 7233);
299			for (i = 0; i < sn; i++)
300				D(i) += fix_mul(SS_(i) + SS_(i + 1), 3633);
301			for (i = 0; i < dn; i++)
302				S(i) = fix_mul(S(i), 5038);	/*5038 */
303			for (i = 0; i < sn; i++)
304				D(i) = fix_mul(D(i), 6659);	/*6660 */
305		}
306	}
307}
308
309static void dwt_encode_stepsize(int stepsize, int numbps, opj_stepsize_t *bandno_stepsize) {
310	int p, n;
311	p = int_floorlog2(stepsize) - 13;
312	n = 11 - int_floorlog2(stepsize);
313	bandno_stepsize->mant = (n < 0 ? stepsize >> -n : stepsize << n) & 0x7ff;
314	bandno_stepsize->expn = numbps - p;
315}
316
317/* 
318==========================================================
319   DWT interface
320==========================================================
321*/
322
323/* <summary>                            */
324/* Forward 5-3 wavelet transform in 2-D. */
325/* </summary>                           */
326void dwt_encode(opj_tcd_tilecomp_t * tilec) {
327	int i, j, k;
328	int *a = NULL;
329	int *aj = NULL;
330	int *bj = NULL;
331	int w, l;
332	
333	w = tilec->x1-tilec->x0;
334	l = tilec->numresolutions-1;
335	a = tilec->data;
336	
337	for (i = 0; i < l; i++) {
338		int rw;			/* width of the resolution level computed                                                           */
339		int rh;			/* height of the resolution level computed                                                          */
340		int rw1;		/* width of the resolution level once lower than computed one                                       */
341		int rh1;		/* height of the resolution level once lower than computed one                                      */
342		int cas_col;	/* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
343		int cas_row;	/* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering   */
344		int dn, sn;
345		
346		rw = tilec->resolutions[l - i].x1 - tilec->resolutions[l - i].x0;
347		rh = tilec->resolutions[l - i].y1 - tilec->resolutions[l - i].y0;
348		rw1= tilec->resolutions[l - i - 1].x1 - tilec->resolutions[l - i - 1].x0;
349		rh1= tilec->resolutions[l - i - 1].y1 - tilec->resolutions[l - i - 1].y0;
350		
351		cas_row = tilec->resolutions[l - i].x0 % 2;
352		cas_col = tilec->resolutions[l - i].y0 % 2;
353        
354		sn = rh1;
355		dn = rh - rh1;
356		bj = (int*)opj_malloc(rh * sizeof(int));
357		for (j = 0; j < rw; j++) {
358			aj = a + j;
359			for (k = 0; k < rh; k++)  bj[k] = aj[k*w];
360			dwt_encode_1(bj, dn, sn, cas_col);
361			dwt_deinterleave_v(bj, aj, dn, sn, w, cas_col);
362		}
363		opj_free(bj);
364		
365		sn = rw1;
366		dn = rw - rw1;
367		bj = (int*)opj_malloc(rw * sizeof(int));
368		for (j = 0; j < rh; j++) {
369			aj = a + j * w;
370			for (k = 0; k < rw; k++)  bj[k] = aj[k];
371			dwt_encode_1(bj, dn, sn, cas_row);
372			dwt_deinterleave_h(bj, aj, dn, sn, cas_row);
373		}
374		opj_free(bj);
375	}
376}
377
378
379/* <summary>                            */
380/* Inverse 5-3 wavelet transform in 2-D. */
381/* </summary>                           */
382void dwt_decode(opj_tcd_tilecomp_t* tilec, int numres) {
383	dwt_decode_tile(tilec, numres, &dwt_decode_1);
384}
385
386
387/* <summary>                          */
388/* Get gain of 5-3 wavelet transform. */
389/* </summary>                         */
390int dwt_getgain(int orient) {
391	if (orient == 0)
392		return 0;
393	if (orient == 1 || orient == 2)
394		return 1;
395	return 2;
396}
397
398/* <summary>                */
399/* Get norm of 5-3 wavelet. */
400/* </summary>               */
401double dwt_getnorm(int level, int orient) {
402	return dwt_norms[orient][level];
403}
404
405/* <summary>                             */
406/* Forward 9-7 wavelet transform in 2-D. */
407/* </summary>                            */
408
409void dwt_encode_real(opj_tcd_tilecomp_t * tilec) {
410	int i, j, k;
411	int *a = NULL;
412	int *aj = NULL;
413	int *bj = NULL;
414	int w, l;
415	
416	w = tilec->x1-tilec->x0;
417	l = tilec->numresolutions-1;
418	a = tilec->data;
419	
420	for (i = 0; i < l; i++) {
421		int rw;			/* width of the resolution level computed                                                     */
422		int rh;			/* height of the resolution level computed                                                    */
423		int rw1;		/* width of the resolution level once lower than computed one                                 */
424		int rh1;		/* height of the resolution level once lower than computed one                                */
425		int cas_col;	/* 0 = non inversion on horizontal filtering 1 = inversion between low-pass and high-pass filtering */
426		int cas_row;	/* 0 = non inversion on vertical filtering 1 = inversion between low-pass and high-pass filtering   */
427		int dn, sn;
428		
429		rw = tilec->resolutions[l - i].x1 - tilec->resolutions[l - i].x0;
430		rh = tilec->resolutions[l - i].y1 - tilec->resolutions[l - i].y0;
431		rw1= tilec->resolutions[l - i - 1].x1 - tilec->resolutions[l - i - 1].x0;
432		rh1= tilec->resolutions[l - i - 1].y1 - tilec->resolutions[l - i - 1].y0;
433		
434		cas_row = tilec->resolutions[l - i].x0 % 2;
435		cas_col = tilec->resolutions[l - i].y0 % 2;
436		
437		sn = rh1;
438		dn = rh - rh1;
439		bj = (int*)opj_malloc(rh * sizeof(int));
440		for (j = 0; j < rw; j++) {
441			aj = a + j;
442			for (k = 0; k < rh; k++)  bj[k] = aj[k*w];
443			dwt_encode_1_real(bj, dn, sn, cas_col);
444			dwt_deinterleave_v(bj, aj, dn, sn, w, cas_col);
445		}
446		opj_free(bj);
447		
448		sn = rw1;
449		dn = rw - rw1;
450		bj = (int*)opj_malloc(rw * sizeof(int));
451		for (j = 0; j < rh; j++) {
452			aj = a + j * w;
453			for (k = 0; k < rw; k++)  bj[k] = aj[k];
454			dwt_encode_1_real(bj, dn, sn, cas_row);
455			dwt_deinterleave_h(bj, aj, dn, sn, cas_row);
456		}
457		opj_free(bj);
458	}
459}
460
461
462/* <summary>                          */
463/* Get gain of 9-7 wavelet transform. */
464/* </summary>                         */
465int dwt_getgain_real(int orient) {
466	(void)orient;
467	return 0;
468}
469
470/* <summary>                */
471/* Get norm of 9-7 wavelet. */
472/* </summary>               */
473double dwt_getnorm_real(int level, int orient) {
474	return dwt_norms_real[orient][level];
475}
476
477void dwt_calc_explicit_stepsizes(opj_tccp_t * tccp, int prec) {
478	int numbands, bandno;
479	numbands = 3 * tccp->numresolutions - 2;
480	for (bandno = 0; bandno < numbands; bandno++) {
481		double stepsize;
482		int resno, level, orient, gain;
483
484		resno = (bandno == 0) ? 0 : ((bandno - 1) / 3 + 1);
485		orient = (bandno == 0) ? 0 : ((bandno - 1) % 3 + 1);
486		level = tccp->numresolutions - 1 - resno;
487		gain = (tccp->qmfbid == 0) ? 0 : ((orient == 0) ? 0 : (((orient == 1) || (orient == 2)) ? 1 : 2));
488		if (tccp->qntsty == J2K_CCP_QNTSTY_NOQNT) {
489			stepsize = 1.0;
490		} else {
491			double norm = dwt_norms_real[orient][level];
492			stepsize = (1 << (gain)) / norm;
493		}
494		dwt_encode_stepsize((int) floor(stepsize * 8192.0), prec + gain, &tccp->stepsizes[bandno]);
495	}
496}
497
498
499/* <summary>                             */
500/* Determine maximum computed resolution level for inverse wavelet transform */
501/* </summary>                            */
502static int dwt_decode_max_resolution(opj_tcd_resolution_t* restrict r, int i) {
503	int mr	= 1;
504	int w;
505	while( --i ) {
506		r++;
507		if( mr < ( w = r->x1 - r->x0 ) )
508			mr = w ;
509		if( mr < ( w = r->y1 - r->y0 ) )
510			mr = w ;
511	}
512	return mr ;
513}
514
515
516/* <summary>                            */
517/* Inverse wavelet transform in 2-D.     */
518/* </summary>                           */
519static void dwt_decode_tile(opj_tcd_tilecomp_t* tilec, int numres, DWT1DFN dwt_1D) {
520	dwt_t h;
521	dwt_t v;
522
523	opj_tcd_resolution_t* tr = tilec->resolutions;
524
525	int rw = tr->x1 - tr->x0;	/* width of the resolution level computed */
526	int rh = tr->y1 - tr->y0;	/* height of the resolution level computed */
527
528	int w = tilec->x1 - tilec->x0;
529
530	h.mem = (int*)opj_aligned_malloc(dwt_decode_max_resolution(tr, numres) * sizeof(int));
531	v.mem = h.mem;
532
533	while( --numres) {
534		int * restrict tiledp = tilec->data;
535		int j;
536
537		++tr;
538		h.sn = rw;
539		v.sn = rh;
540
541		rw = tr->x1 - tr->x0;
542		rh = tr->y1 - tr->y0;
543
544		h.dn = rw - h.sn;
545		h.cas = tr->x0 % 2;
546
547		for(j = 0; j < rh; ++j) {
548			dwt_interleave_h(&h, &tiledp[j*w]);
549			(dwt_1D)(&h);
550			memcpy(&tiledp[j*w], h.mem, rw * sizeof(int));
551		}
552
553		v.dn = rh - v.sn;
554		v.cas = tr->y0 % 2;
555
556		for(j = 0; j < rw; ++j){
557			int k;
558			dwt_interleave_v(&v, &tiledp[j], w);
559			(dwt_1D)(&v);
560			for(k = 0; k < rh; ++k) {
561				tiledp[k * w + j] = v.mem[k];
562			}
563		}
564	}
565	opj_aligned_free(h.mem);
566}
567
568static void v4dwt_interleave_h(v4dwt_t* restrict w, float* restrict a, int x, int size){
569	float* restrict bi = (float*) (w->wavelet + w->cas);
570	int count = w->sn;
571	int i, k;
572	for(k = 0; k < 2; ++k){
573		if (count + 3 * x < size && ((size_t) a & 0x0f) == 0 && ((size_t) bi & 0x0f) == 0 && (x & 0x0f) == 0) {
574			/* Fast code path */
575			for(i = 0; i < count; ++i){
576				int j = i;
577				bi[i*8    ] = a[j];
578				j += x;
579				bi[i*8 + 1] = a[j];
580				j += x;
581				bi[i*8 + 2] = a[j];
582				j += x;
583				bi[i*8 + 3] = a[j];
584			}
585		} else {
586			/* Slow code path */
587		for(i = 0; i < count; ++i){
588			int j = i;
589			bi[i*8    ] = a[j];
590			j += x;
591			if(j > size) continue;
592			bi[i*8 + 1] = a[j];
593			j += x;
594			if(j > size) continue;
595			bi[i*8 + 2] = a[j];
596			j += x;
597			if(j > size) continue;
598			bi[i*8 + 3] = a[j];
599		}
600		}
601		bi = (float*) (w->wavelet + 1 - w->cas);
602		a += w->sn;
603		size -= w->sn;
604		count = w->dn;
605	}
606}
607
608static void v4dwt_interleave_v(v4dwt_t* restrict v , float* restrict a , int x){
609	v4* restrict bi = v->wavelet + v->cas;
610	int i;
611	for(i = 0; i < v->sn; ++i){
612		memcpy(&bi[i*2], &a[i*x], 4 * sizeof(float));
613	}
614	a += v->sn * x;
615	bi = v->wavelet + 1 - v->cas;
616	for(i = 0; i < v->dn; ++i){
617		memcpy(&bi[i*2], &a[i*x], 4 * sizeof(float));
618	}
619}
620
621#ifdef __SSE__
622
623static void v4dwt_decode_step1_sse(v4* w, int count, const __m128 c){
624	__m128* restrict vw = (__m128*) w;
625	int i;
626	/* 4x unrolled loop */
627	for(i = 0; i < count >> 2; ++i){
628		*vw = _mm_mul_ps(*vw, c);
629		vw += 2;
630		*vw = _mm_mul_ps(*vw, c);
631		vw += 2;
632		*vw = _mm_mul_ps(*vw, c);
633		vw += 2;
634		*vw = _mm_mul_ps(*vw, c);
635		vw += 2;
636	}
637	count &= 3;
638	for(i = 0; i < count; ++i){
639		*vw = _mm_mul_ps(*vw, c);
640		vw += 2;
641	}
642}
643
644static void v4dwt_decode_step2_sse(v4* l, v4* w, int k, int m, __m128 c){
645	__m128* restrict vl = (__m128*) l;
646	__m128* restrict vw = (__m128*) w;
647	int i;
648	__m128 tmp1, tmp2, tmp3;
649	tmp1 = vl[0];
650	for(i = 0; i < m; ++i){
651		tmp2 = vw[-1];
652		tmp3 = vw[ 0];
653		vw[-1] = _mm_add_ps(tmp2, _mm_mul_ps(_mm_add_ps(tmp1, tmp3), c));
654		tmp1 = tmp3;
655		vw += 2;
656	}
657	vl = vw - 2;
658	if(m >= k){
659		return;
660	}
661	c = _mm_add_ps(c, c);
662	c = _mm_mul_ps(c, vl[0]);
663	for(; m < k; ++m){
664		__m128 tmp = vw[-1];
665		vw[-1] = _mm_add_ps(tmp, c);
666		vw += 2;
667	}
668}
669
670#else
671
672static void v4dwt_decode_step1(v4* w, int count, const float c){
673	float* restrict fw = (float*) w;
674	int i;
675	for(i = 0; i < count; ++i){
676		float tmp1 = fw[i*8    ];
677		float tmp2 = fw[i*8 + 1];
678		float tmp3 = fw[i*8 + 2];
679		float tmp4 = fw[i*8 + 3];
680		fw[i*8    ] = tmp1 * c;
681		fw[i*8 + 1] = tmp2 * c;
682		fw[i*8 + 2] = tmp3 * c;
683		fw[i*8 + 3] = tmp4 * c;
684	}
685}
686
687static void v4dwt_decode_step2(v4* l, v4* w, int k, int m, float c){
688	float* restrict fl = (float*) l;
689	float* restrict fw = (float*) w;
690	int i;
691	for(i = 0; i < m; ++i){
692		float tmp1_1 = fl[0];
693		float tmp1_2 = fl[1];
694		float tmp1_3 = fl[2];
695		float tmp1_4 = fl[3];
696		float tmp2_1 = fw[-4];
697		float tmp2_2 = fw[-3];
698		float tmp2_3 = fw[-2];
699		float tmp2_4 = fw[-1];
700		float tmp3_1 = fw[0];
701		float tmp3_2 = fw[1];
702		float tmp3_3 = fw[2];
703		float tmp3_4 = fw[3];
704		fw[-4] = tmp2_1 + ((tmp1_1 + tmp3_1) * c);
705		fw[-3] = tmp2_2 + ((tmp1_2 + tmp3_2) * c);
706		fw[-2] = tmp2_3 + ((tmp1_3 + tmp3_3) * c);
707		fw[-1] = tmp2_4 + ((tmp1_4 + tmp3_4) * c);
708		fl = fw;
709		fw += 8;
710	}
711	if(m < k){
712		float c1;
713		float c2;
714		float c3;
715		float c4;
716		c += c;
717		c1 = fl[0] * c;
718		c2 = fl[1] * c;
719		c3 = fl[2] * c;
720		c4 = fl[3] * c;
721		for(; m < k; ++m){
722			float tmp1 = fw[-4];
723			float tmp2 = fw[-3];
724			float tmp3 = fw[-2];
725			float tmp4 = fw[-1];
726			fw[-4] = tmp1 + c1;
727			fw[-3] = tmp2 + c2;
728			fw[-2] = tmp3 + c3;
729			fw[-1] = tmp4 + c4;
730			fw += 8;
731		}
732	}
733}
734
735#endif
736
737/* <summary>                             */
738/* Inverse 9-7 wavelet transform in 1-D. */
739/* </summary>                            */
740static void v4dwt_decode(v4dwt_t* restrict dwt){
741	int a, b;
742	if(dwt->cas == 0) {
743		if(!((dwt->dn > 0) || (dwt->sn > 1))){
744			return;
745		}
746		a = 0;
747		b = 1;
748	}else{
749		if(!((dwt->sn > 0) || (dwt->dn > 1))) {
750			return;
751		}
752		a = 1;
753		b = 0;
754	}
755#ifdef __SSE__
756	v4dwt_decode_step1_sse(dwt->wavelet+a, dwt->sn, _mm_set1_ps(K));
757	v4dwt_decode_step1_sse(dwt->wavelet+b, dwt->dn, _mm_set1_ps(c13318));
758	v4dwt_decode_step2_sse(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), _mm_set1_ps(dwt_delta));
759	v4dwt_decode_step2_sse(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), _mm_set1_ps(dwt_gamma));
760	v4dwt_decode_step2_sse(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), _mm_set1_ps(dwt_beta));
761	v4dwt_decode_step2_sse(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), _mm_set1_ps(dwt_alpha));
762#else
763	v4dwt_decode_step1(dwt->wavelet+a, dwt->sn, K);
764	v4dwt_decode_step1(dwt->wavelet+b, dwt->dn, c13318);
765	v4dwt_decode_step2(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), dwt_delta);
766	v4dwt_decode_step2(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), dwt_gamma);
767	v4dwt_decode_step2(dwt->wavelet+b, dwt->wavelet+a+1, dwt->sn, int_min(dwt->sn, dwt->dn-a), dwt_beta);
768	v4dwt_decode_step2(dwt->wavelet+a, dwt->wavelet+b+1, dwt->dn, int_min(dwt->dn, dwt->sn-b), dwt_alpha);
769#endif
770}
771
772/* <summary>                             */
773/* Inverse 9-7 wavelet transform in 2-D. */
774/* </summary>                            */
775void dwt_decode_real(opj_tcd_tilecomp_t* restrict tilec, int numres){
776	v4dwt_t h;
777	v4dwt_t v;
778
779	opj_tcd_resolution_t* res = tilec->resolutions;
780
781	int rw = res->x1 - res->x0;	/* width of the resolution level computed */
782	int rh = res->y1 - res->y0;	/* height of the resolution level computed */
783
784	int w = tilec->x1 - tilec->x0;
785
786	h.wavelet = (v4*) opj_aligned_malloc((dwt_decode_max_resolution(res, numres)+5) * sizeof(v4));
787	v.wavelet = h.wavelet;
788
789	while( --numres) {
790		float * restrict aj = (float*) tilec->data;
791		int bufsize = (tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0);
792		int j;
793
794		h.sn = rw;
795		v.sn = rh;
796
797		++res;
798
799		rw = res->x1 - res->x0;	/* width of the resolution level computed */
800		rh = res->y1 - res->y0;	/* height of the resolution level computed */
801
802		h.dn = rw - h.sn;
803		h.cas = res->x0 % 2;
804
805		for(j = rh; j > 3; j -= 4){
806			int k;
807			v4dwt_interleave_h(&h, aj, w, bufsize);
808			v4dwt_decode(&h);
809				for(k = rw; --k >= 0;){
810					aj[k    ] = h.wavelet[k].f[0];
811					aj[k+w  ] = h.wavelet[k].f[1];
812					aj[k+w*2] = h.wavelet[k].f[2];
813					aj[k+w*3] = h.wavelet[k].f[3];
814				}
815			aj += w*4;
816			bufsize -= w*4;
817		}
818		if (rh & 0x03) {
819				int k;
820			j = rh & 0x03;
821			v4dwt_interleave_h(&h, aj, w, bufsize);
822			v4dwt_decode(&h);
823				for(k = rw; --k >= 0;){
824					switch(j) {
825						case 3: aj[k+w*2] = h.wavelet[k].f[2];
826						case 2: aj[k+w  ] = h.wavelet[k].f[1];
827						case 1: aj[k    ] = h.wavelet[k].f[0];
828					}
829				}
830			}
831
832		v.dn = rh - v.sn;
833		v.cas = res->y0 % 2;
834
835		aj = (float*) tilec->data;
836		for(j = rw; j > 3; j -= 4){
837			int k;
838			v4dwt_interleave_v(&v, aj, w);
839			v4dwt_decode(&v);
840				for(k = 0; k < rh; ++k){
841					memcpy(&aj[k*w], &v.wavelet[k], 4 * sizeof(float));
842				}
843			aj += 4;
844		}
845		if (rw & 0x03){
846				int k;
847			j = rw & 0x03;
848			v4dwt_interleave_v(&v, aj, w);
849			v4dwt_decode(&v);
850				for(k = 0; k < rh; ++k){
851					memcpy(&aj[k*w], &v.wavelet[k], j * sizeof(float));
852				}
853			}
854	}
855
856	opj_aligned_free(h.wavelet);
857}
858